ABSTRACT
Empirical likelihood was motivated by independent identically distributed data. As Theorem 4.1 shows, the requirement for identically distributed data can be relaxed. When the observations are dependent, then this usually has to be accounted for in constructing confidence regions and tests. The ways of handling dependent data with empirical likelihood parallel the methods from parametric likelihood and the bootstrap. Failure to account for dependence among the data can destroy the coverage properties of confidence regions.
